Radeon ROCm 3.5 Released With New Features But Still No Navi Support

[walterav]

Hold on, I screwed up... I was confusing 280X (Tahiti, gfx6) with 285/380 (Tonga, early gfx8). It's Tonga that was "on the edge" - Tahiti doesn't have the MEC block that ROCm is built over. Sorry about that.

Don't you mean GFX7 Hawaii 390x thats also in the readme.md, however its support is broken since rocm 2.0, not even unofficial supported.

In these specs I see alot of "mec" for gfx7 is that the MEC block that is needed like gfx8?:
https://elixir.bootlin.com/linux/lat...gpu/gfx_v7_0.c

[agd5f]

gfx7 was the first family of chips that supported the necessary features for ROCm (separate MEC engines, hardware scheduler, compute pre-emption, etc.). All newer versions of the gfx block had that functionality (gfx8, 9, 10).

[bridgman]

Don't you mean GFX7 Hawaii 390x thats also in the readme.md, however its support is broken since rocm 2.0, not even unofficial supported.

In these specs I see alot of "mec" for gfx7 is that the MEC block that is needed like gfx8?:
https://elixir.bootlin.com/linux/lat...gpu/gfx_v7_0.c

Hawaii (gfx7) had an early version of MEC with a fixed microcode store in the MEC block, while the later gfx8 parts could support larger microcode images by executing directly out of VRAM via an instruction cache. I had not heard about Hawaii support being broken until today but will take a look. It was always a challenge to fit even stripped-down ROCm functionality into the fixed microcode store so it's possible that we just outgrew it.

Tonga was the first of the gfx8 parts and so the first to be able to run larger microcode images in MEC. IIRC it had some HW issues in the MEC block as well (it went to production before the ROCm stack was developed) that did not affect graphics or the AMDGPU-PRO OpenCL paths but which did get in the way when we tried running the ROCm stack on it.

Fiji, the next chip after Tonga, was the first with what I would consider full and working HW support.

[pal666]

Yes, it absolutely is... run Nvidia if you have anything related to GPU Compute! Its not one click setup, but it works and thats probably the key feature

moron, does novideo work on navi?

[pal666]

Fuck NVIDIA, right?

Oh wait, NVIDIA does fully support OpenCL 1.2

moron, is it so easy to confuse opencl 2.2 with opencl 1.2?

[pal666]

July 2019 and now a full year later

you can't produce one sentence without lies, can you?

[Djhg2000]

Well I'm using an RX 5500 XT in my Debian Sid system using Linux 5.7 and it's working just fine except for ROCm. With some older kernels I had a few hiccups but it's fine now.

[JustRob]

According to AMD ROCm will support HPC but the focus will be on the Instinct line over supporting every GPU.
There's a lot that goes into supporting each card and the effort can't be made for every one all at once.

[walterav]

...
I had not heard about Hawaii support being broken until today but will take a look. It was always a challenge to fit even stripped-down ROCm functionality into the fixed microcode store so it's possible that we just outgrew it.
...

Although the hardware is offcourse dated nowadays, it would be great if it can have another look even if its only to correct the readme.md to get GFX7 removed to solve the confusion.

https://github.com/RadeonOpenCompute/ROCm/issues/871

[tildearrow]

It's a good card. Your biggest mistake is probably not doing your homework before buying

I would like to ask. Do you do any of these with your AMD card?:

- play AAA games? (if so, did it hang for you?)
- record the desktop using VA-API? (if so, did it hang for you?)
- leave it on (without logging out) for more than 2 days? (if so, did it hang for you?)